Robot designs
Dan Taylor
Posts: 207
Hi,
· I am just looking for some robot designs. If you have any documents or websites they will be gladly accepted.
I'm looking for a 2 wheeled robot·with probably a 3rd wheel for balance.
Thanks!!
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Dan Taylor
· I am just looking for some robot designs. If you have any documents or websites they will be gladly accepted.
I'm looking for a 2 wheeled robot·with probably a 3rd wheel for balance.
Thanks!!
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Dan Taylor
Comments
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
- Stephen
Thanks·Franklin·for the link.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Dan Taylor
How would you add in more voltage. Is there an easier way than adding more batteries. like combining the VIN with the VDD or something?
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Dan Taylor
Spoiler: I'm no fan of·differential drive platforms, and my comments to that end·may draw spirited responses from the forum.
Don't just cop out and build another differential drive robot that never goes straight. No one ever talks about how poorly differential drive robots navigate, so make it a point to ask 'em sometime. Unless you have multiple wheel watchers (encoders)·and dedicated processors to reduce (never eliminate) steering errors, your robot will·rarely do the same thing twice.
Diff drive robots are the easiest to build, and the most difficult to get anything useful out of. Just because it's quick & easy to stick two continuous-rotation servos on a·chassis doesn't "make it right". There are plenty of bad examples out there, no point in repeating the mistakes of the past. Many people assume that jamming lots of sonar & infrared sensors on a diff-drive bot, plus some "slick software" will somehow make up for a fundamentally flawed chassis design. It ain't happened yet.
IMHO, if·a robot isn't accurate enough to dead-reckon it's way·15 feet out , make a few turns, and come back to within a foot of its starting position, it's not a great candidate to expect much repeatability from.
I suggest you do your homework on chassis design and pick one that makes sense for your application. Consider·tricycle or car-drive steering. Don't be in a hurry to slap your robot together.·The steering system is the most mechanically critical item in your whole chassis. You must have tight steering pivots, no slop, no flex, and an ordinary servo won't cut it for strength or repeatability on a robot that's large enough to be useful. Some custom motor-driven geartrain with feedback will give far better results than a small servo.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
·"If you build it, they will come."
Post Edited (erco) : 6/1/2008 4:34:27 AM GMT
For someone like myself who knows just enough mechanics to be dangerous,
(simple levers still fascinate me.) Building a small differential drive type robot
(Which is exactly what I'm doing) is an important step in my robotics education.
No, I don't expect it to be accurate for the reasons given, but I do expect to
learn a *LOT* about basic design, servo control, programming, etc.
Yeah, I know.. I've heard the statement "There goes the neighborhood" before. <SMIRK>
Newbie/OBC
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
New to the Propeller?
Getting started with the Protoboard? - Propeller Cookbook 1.4
Updates to the Cookbook are now posted to: Propeller.warrantyvoid.us
Got an SD card? - PropDOS
Need a part? Got spare electronics? - The Electronics Exchange
Attached is a graph showing wheel speed (read from the Parallax BOE-Bot Digital Encoder) vs. servo supply voltage at a PULSOUT value (BS2) of 850. The highest speed I measured could be obtained with the Parallax BOE-Boost and five alkaline batteries, while setting the servo supply jumper to Vin.
Erco,
I respectfully disagree with your take on differential drive systems. (I assume you added a bit of hyperbole to spur discussion, and I'm always an easy mark for taking the bait.
Coordinated wheel motion is pretty simple, even using a BOE-Bot equipped with encoders. In fact, with the downloadable coordinated motion program, and running a straight-line course, you can hold one wheel to slow and stall it, and the other will slow and stall at the same rate — as though they were joined by a rigid shaft.
-Phil
You did take my bait on this one. I'm an ME, a machinist,·and a simplicity freak (the KISS principle). I like your comparison between a diffdrive & solid axle, that's exactly the road I'm going down. It's my own personal conviction that an affordable·consumer·robot must rely heavily on accurate dead reckoning by sheer mechanical repeatability, using only·minimal sensor input to verify position.·I'll agree that properly-used encoders can somewhat approach (but never equal) a rigid axle. There are·issues of encoder resolution, wheel "catch-up", etc. Feedback loops work to take action only after an error (i.e., one wheel ahead of the other)·has been detected. By that time, the damage has been done and any momentary variation in·robot's angular heading will integrate into a sizeable cumulative positional error.
Beyond that, my learned opponent (grin), I·will still cling to my firm beliefs that mechanical construction details are the most critical, and yet most overlooked, in the building of any robot. Whether diffdrive or otherwise, most builders are forced to use existing mechanical parts. Wheels, transmissions, motors, etc. Many wheels aren't perfectly circular.·David Cook's·robot building for beginners (while a fine book) shows an illustration of an encoder disk with the hole punched off-center from the radial pattern.·Either of these, or backlash, chassis flex, or tolerance variations·will·introduce non-linearities·in an encoder system and contribute to poor repeatability.
Long story short, I'd like to see more people·getting their hands dirty in a machine shop. Good mechanical building is almost a lost art, and most people would never consider making their own wheels or transmissions. Actually, it is fun and liberating to make something that you designed and/or can't buy. Turn (or true up) your own wheels in a lathe. Make a delrin bushing insert for your stamped aluminum chassis. IMHO, until you have adjusted gear backlash perfectly·by making YET ANOTHER gearbox, you can't have a full appreciation of the spectrum of talents it takes to be a roboticist. One of my motivations for my initial post here was simply to say, don't settle for commonly available parts. If you think up a better way to do something, build it. A lot of community colleges have machine shop classes. Sign up now, you'll love it!
Of course, even rigid axle chasses will have·some repeatability issues, (wheel slip, etc.) so·I'll quit before I alienate everyone·in this forum. I can tell you·that I am working hard on two·exciting projects that are breakthrough robot navigation platforms heading for the patent office. I·have an inexpensive·new consumer robot·I'm finalizing the design on that every man in America will want, no NEED. And right behind that is·the world's simplest·diffdrive variant that locks the axle solid for straight-line travel.··Same action as·a dual-differential drive, but with 90% fewer parts; I told you I was a simplicity freak.
Cheers, PhiPi and others: now that I've drawn your ire, I'll look for and welcome your alternate and well-informed viewpoints.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
·"If you build it, they will come."
a good try at getting the most from your home built bot is by testing muliple parts for closest exact match's like any resistors or capacitors used in wheel motors,servo's a variable power supply and function genorator &good ohmmeter will help ,in most places you dont need this tight tollerance but in drive motor circuits it is deffinatly needed ,also note they dont have to be the exact value labled on the part but exactly the same as the one on the other drive/wheel motor. if these things are tested to exact match and sensors are used the outcome is much more close but still a flaw will exist and it can be adjusted by software wich will get you even closer to acuracy. new to our local hospitol here are surgical robots which are no different form what we build. knowing this ,if a robot can perform surgery and a patient live I'm thinking it is also possible that the flaw can be reduced to an acceptable tollerance and perform in our public socioty. the question isnt how to make it acurate as much as what amount of inacuracy will be acceptable for your application ,and how to refine it to that point.
I can certainly appreciate your passion for mechanical soundness, and I can tell from these and other posts that your skills as a designer and educator run deep. As I'm sure you do too, I decry the loss of basic tool skills among younger folk. (When did machine shop cease being a required course in middle school?) However, as an entry requirement for robotics, what you propose creates quite a bump in the road. Imagine where Parallax would be if they required their customers to build gearboxes for the BOE-Bot or to true its wheels on a lathe. The geniius of feedback systems is that they can wring fairly precise performance from mechanisms that are built to loose tolerances. Your comment about having to commit an error to detect one is certainly true. But how big of an error? With a fine-enough encoder disk, for example, odometry errors can be minimized. Even a rigid shaft can experience torsional deflection on rough terrain, after all.
Precisely-built clockwork mechanisms are are pleasure to behold and a joy to own. But let's face it, they're also expensive to produce. One measure of any good mechanical design is the ability to perform to specifications that exceed those of its individual parts. A perfect example of this was the IBM model 75, "Selectronic" typewriter I once owned. Its roughly 1000 mechanical components consisted mostly of bent metal stampings and zinc die-castings, with a fair number of springs thrown in for good measure. There was barely a machined part in the thing. Moreover, its screw holes were extra large to facilitate rapid assembly. It was, in equal parts, mechanical marvel and monstrosity: the pinnacle achievement of IBM's typewriter division. But despite its lack of tightly-toleranced components, it did its design task well: depositing crisp type, precisely positioned on the page — even proportionally spaced, given the right typeball.
That electronics is able to augment this performance-to-precision ratio is, to me at least, a given. What result are devices more organic and less rigid in nature. When properly implemented, the options for responding to environmental variations are multiplied over those of even the best-crafted, but mechanically rigid, marvels. That I'm able to pick up a pen, for example, and sign my name legibly, is not testament to finely-machined finger joints but, rather, to the collection of nerves feeding back to the muscles that move those joints. And even these were not designed for the task but evolved for malleability, so they could be trained to write with a pen, play an instrument, operate a jeweler's lathe, or craft a flint spearhead. It's to this ideal that roboticists should aspire, in my not-so-humble opinion. Design with loose tolerances and loose objectives, but with enough underlying fexibility that desired, complex behaviors can emerge from simple, imprecise structures.
The Victorian Age produced some wonderful mechanical automatons — beautiful in their execution and breathtaking for their complexity. And, as you must also, I appreciate these marvelous devices for the watchmaker's art that they represent. But the new science of complexity theory and emergent behavior has stolen the show, in my mind, and will be the basis for robotic advances in this century. I'm talking Mark Tilden-style robotics, but augmented by steroids (i.e. Propeller-grade microcontrollers). Be that as it may, the tantalizing hints you've dangled regarding your new, soon-to-be-patented drive mechanisms intrigue me, and I look forward to seeing more when the time is right!
Now I'm sure my comments have provided plenty of grist for further discussion. I'm rather enjoying this, you know?
-Phil
__________________________
Dan,
My apologies if this discussion has veered off from your original topic. If you feel that your thread has been hijacked, please say so, and we can copy these posts to a new one.
-Phil
I talked to the guy who made Cye a few years ago, I can't remember his name. But it was mostly his own personal venture to make and market this robot, and he did a tremendous job all-around on designing and building the robot and the software. There is an amazing·GUI to map out your house with, and it is pretty accurate. Best of all, there·are NO sensors on this robot other than the encoders. It is 100% blind. It verifies its position periodically by squaring itself up with a wall or corner. It can also dock with its charger by dead reckoning. It uses the encoders like bump switches, so it knows when it hits something since those spiked wheels stop in a hurry.
I would venture to say that may be the most accurate dead-reckoning consumer robot ever built. It did still slowly accumulate errors, but had provisions to correct itself along the way. BTW, I'll jump to this great bot's defense and admit that much of that error was causes by the deep pile carpet in my house when I was testing the robot. That carpet introduced a lot of ground slip.
The refreshing change here is that·CYE was designed from the start as a real-world robot for use in a real house, and LOTS of thought, testing, and redesign went into this. The geometry of the robot (large wheels·and rectangular chassis design) lends itself to optimum dead-reckoning ability, and dedicated high-speed encoders and multiple processors had·this sound electromechanical foundation to enhance, not·compensate for. In contrast, most modern diffdrive robots·have round chasses and·small wheels, a big step backward IMHO. Seems like CYE would be a diffdrive chassis worthy of·studying and emulating. Of course, it would take several single-thread Stamps or a Propeller chip to keep up with Cye's dedicated·electronics.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
·"If you build it, they will come."
You and I are certainly enjoying this banter, even if no one else is. And no, I'm not suggesting that Parallax change anything. Quite the contrary, I dearly love my Stamps, Boe-Bot and Scribblers! They are the best introductory tools I can and do recommend. My thoughts are "what after Boe-Bot", and that's pretty much what Dan's reply inferred. There are certain compromises and limitations to any robot design, and I just like to keep progress moving forward. Somebody here is gonna be the Bill Gates of consumer robotics, and I just hope to be able to say "I knew him way back when"...
BTW, I have great memories of a bicycle trip through Port Townsend (your location?) a few years back. I started in Vancouver and pedaled home to LA. After my derailleur broke off in the middle of the Hoh Rain forest, I hitched back to Forks, where I met the nicest folks in the world who welded my bike up and got me back on the road. So now I know you Washingtonians are nice, helpful AND smart!
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
·"If you build it, they will come."
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Whit+
"We keep moving forward, opening new doors, and doing new things, because we're curious and curiosity keeps leading us down new paths." - Walt Disney
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
·"If you build it, they will come."
Clearly, differentially steered platforms can have incredibly accurate dead reckoning -- it's been done many times. See esp. some of David Anderson's hobby efforts (www.geology.smu.edu/~dpa-www/robo/jbot/index.html is a very slick outdoor 'bot with great long-range dead-reckoning; he also has some useful articles on odometry in general).
Additionally, the UMBMark provides an extremely useful and precise methodology for correcting and calibrating dead-reckoning platforms, e.g., correcting for differences in true axle length, true wheel circumference, etc. It's a crucial paper for any kind of dead-reckoning on any kind of platform: www-personal.engin.umich.edu/~johannb/Papers/umbmark.pdf.
However, "ackerman" steering is *vastly* more stable at high speeds, and can bear heavy loads in a better way.
So for a 'bot moving at a fast human walk speed (or slower), give me differential steering. For high speeds, give me a 4 wheeled auto type chassis w/ackerman steering.
In either case, odometry for dead-reckoning will be problematic without overlapping sensors (e.g. odometry + inertial measurement + gyroscope + possibly GPS + landmark reckoning). Any kind of platform will lose traction at some point. Cars skid out on tight turns. Etc.
A final P.S. -- personally, I think the steered "trike" is the worst of both worlds -- it lacks the stability of 4-wheeled ackerman platforms at high speed, and lacks the maneuverability and simplicity of differentially-steered platforms. Uh, like my Hero-1.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
When the going gets weird, the weird turn pro. -- HST
1uffakind.com/robots/povBitMapBuilder.php
1uffakind.com/robots/resistorLadder.php
all the talk about dead reckoning and pin point accuracy. There
are certainly applications for this kind of accuracy. But here's
a thought - why not replace the dead reckoning mechanisms
with AI. When humans approach the door, they are never
standing in exactly the same place, yet the intelligence to
place a hand exactly on the door knob, is successful. Perhaps
we can look at other ways to gain accuracy in repeatability,
even if the motions to get there are approximate.
humanoido
·If you really want to build a robot and not just put a kit together then see my robot at:
http://www.instructables.com/contest/robotcontest/?show=ENTRIES
A Practical Robot
by
Al1970
I call it a practical robot for a number of reasons. It can be make using every day tools that most people who do work around the house would have. By using many surplus items the cost is kept down. The robot's arm can lift over a 2 lb. object from the floor to 3 ft. 4 in. in the air, so the arm can put objects onto tables. So if you are tired of reading about robots that can only lift a ping pong ball a few inchs in the air then read my tutorial and VOTE FOR ME.